My goal in this project is to understand which features of a country’s government are correlated with greater happiness. I will use visual and predictive techniques to conduct my analysis.
# A tibble: 2,132 × 18
country year happiness happiness_sd log_gdp_per_capita social_support
<chr> <dbl> <dbl> <dbl> <dbl> <dbl>
1 Afghanistan 2005 NA NA NA NA
2 Afghanistan 2006 NA NA NA NA
3 Afghanistan 2007 NA NA NA NA
4 Afghanistan 2008 3.72 1.77 7.17 0.451
5 Afghanistan 2009 4.40 1.72 7.33 0.552
6 Afghanistan 2010 4.76 1.88 7.39 0.539
7 Afghanistan 2011 3.83 1.79 7.42 0.521
8 Afghanistan 2012 3.78 1.80 7.52 0.521
9 Afghanistan 2013 3.57 1.22 7.50 0.484
10 Afghanistan 2014 3.13 1.40 7.48 0.526
# ℹ 2,122 more rows
# ℹ 12 more variables: life_expectancy <dbl>, freedom_choices <dbl>,
# generosity <dbl>, corruption <dbl>, positive_affect <dbl>,
# negative_affect <dbl>, government_confidence <dbl>,
# democratic_quality <dbl>, delivery_quality <dbl>, gini_gallup <dbl>,
# gini_world_bank <dbl>, gini_world_bank_average <dbl>
Making a plan for predictability
My hope is that my findings will be relevant over time. Since I plan to just use the 2017 data to conduct my analysis, and I will use the 2016 data to evaluate the predictability. (No split is needed).
Cleaning and preprocessing the data for analysis
The following code creates a version of the cleaned/pre-processed data in which we:
Drop any countries that are missing the 2017 happiness score
# filter to 2017happiness_2017 <- happiness_clean |>select(country, year, corruption, gini, happiness) |>drop_na() |>filter(year ==2017)# scatterplot of corruption vs happinessgg_corruption <-ggplot(happiness_2017, aes(x = corruption, y = happiness)) +geom_point() +geom_smooth(method ="lm") +ggtitle(paste0("cor=", round(cor(happiness_2017$corruption, happiness_2017$happiness), 2)))# scatterplot of gini index vs happinessgg_gini <-ggplot(happiness_2017, aes(x = gini, y = happiness)) +geom_point() +geom_smooth(method ="lm") +ggtitle(paste0("cor=", round(cor(happiness_2017$gini, happiness_2017$happiness), 2)))# make a grid of plots using + from patchwork librarygg_corruption + gg_gini
`geom_smooth()` using formula = 'y ~ x'
`geom_smooth()` using formula = 'y ~ x'
The following code creates a function for this analysis with the “year” as a an argument to make predictability analysis easy.
happinessAnalysis <-function(happiness_clean, .year =2017) {# filter to the selected year happiness_year <- happiness_clean |>select(country, year, corruption, gini, happiness) |>drop_na() |>filter(year == .year)# scatterplot of corruption vs happiness gg_corruption <-ggplot(happiness_year, aes(x = corruption, y = happiness)) +geom_point() +geom_smooth(method ="lm") +ggtitle(paste0(.year, " | cor=", round(cor(happiness_year$corruption, happiness_year$happiness), 2)))# scatterplot of gini index vs happiness gg_gini <-ggplot(happiness_year, aes(x = gini, y = happiness)) +geom_point() +geom_smooth(method ="lm") +ggtitle(paste0(.year, " | cor=", round(cor(happiness_year$gini, happiness_year$happiness), 2)))# make a grid of plots using + from patchwork library gg_corruption + gg_gini }
happinessAnalysis(happiness_clean, .year =2015)
`geom_smooth()` using formula = 'y ~ x'
`geom_smooth()` using formula = 'y ~ x'
Predictability analysis
happinessAnalysis(happiness_clean, .year =2016)
`geom_smooth()` using formula = 'y ~ x'
`geom_smooth()` using formula = 'y ~ x'
happinessAnalysis(happiness_clean, .year =2017)
`geom_smooth()` using formula = 'y ~ x'
`geom_smooth()` using formula = 'y ~ x'
Stability analysis
Judgment calls
Let’s consider the stability of our analysis to our judgment calls
`geom_smooth()` using formula = 'y ~ x'
`geom_smooth()` using formula = 'y ~ x'
# using interpolation to impute rather than forward fillhappiness_clean_interpolation <-prepareData(happiness_orig, drop_countries_missing_2017_happiness =FALSE,gini ="gallup", imputation ="interpolation")happinessAnalysis(happiness_clean_interpolation, .year =2017)
`geom_smooth()` using formula = 'y ~ x'
`geom_smooth()` using formula = 'y ~ x'
# using world bank gini measurehappiness_clean_wb <-prepareData(happiness_orig, drop_countries_missing_2017_happiness =FALSE,gini ="world bank", imputation ="ffill")happinessAnalysis(happiness_clean_wb, .year =2017)
`geom_smooth()` using formula = 'y ~ x'
`geom_smooth()` using formula = 'y ~ x'
# using world bank gini measure with interpolationhappiness_clean_wb_interpolation <-prepareData(happiness_orig, drop_countries_missing_2017_happiness =FALSE,gini ="world bank", imputation ="interpolation")happinessAnalysis(happiness_clean_wb_interpolation, .year =2017)
`geom_smooth()` using formula = 'y ~ x'
`geom_smooth()` using formula = 'y ~ x'
Data perturbations
Since all countries are already included in our dataset, sampling-based perturbations do not really make sense because there is no sampling uncertainty at the country-level (there is at the respondent level within each country, but we don’t have access to that data).
However, the actual average variable value reported for each country has some uncertainty. While we have the standard deviation for each country’s happiness score, we do not have this value for the other values.
# create 10 perturbed versions of the dataperturbData <-function(happiness_data) { happiness_data_perturbed <- happiness_data |># for each row:drop_na(happiness) |>rowwise() |># add a random normally-distributed value to each happiness score within 1 SDmutate(happiness = happiness +rnorm(1, mean =0, sd = happiness_sd)) |>ungroup()return(happiness_data_perturbed)}
# original results:happinessAnalysis(happiness_clean, .year =2017)
`geom_smooth()` using formula = 'y ~ x'
`geom_smooth()` using formula = 'y ~ x'
# perturbed versions of the resultshappinessAnalysis(perturbData(happiness_clean), .year =2017)
`geom_smooth()` using formula = 'y ~ x'
`geom_smooth()` using formula = 'y ~ x'
